The demand for embedded nonvolatile memory (NVM) in integrated circuits has grown steadily over the past decade. Desirable characteristics of embedded NVM include low cost, low power, high speed, and high reliability (data retention and program/erase cycling endurance). NVM may be embedded in various integrated circuit (IC) technologies such as, for example, the widely used Complementary Metal Oxide Semiconductor (CMOS) technology. Some embedded CMOS applications include, for example, storing: (1) chip serial numbers, (2) configuration information in ASICs (Application Specific Integrated Circuits), (3) product data in radio frequency identification integrated circuits, (4) code or data in embedded microcontrollers, and (5) analog trim information.
A major barrier for using embedded NVM is cost. An IC fabricator typically requires additional processing steps to manufacture NVM storage transistors. For example, IC fabricators sometimes use two layers of polysilicon for the gate of an NVM storage transistor, rather than one layer as in standard CMOS technology. The additional fabrication step increases the total cost of the IC. Typical embedded EEPROM (electrically erasable programmable read only memory) or Flash NVM uses nFET (n-channel field effect transistor) storage transistors. To ensure charge retention in nFETs, the IC fabricator typically uses a thicker gate oxide than is found in logic transistors, again increasing cost.
To reduce the costs and added complexities of embedding NVM in integrated circuits, efforts have been made to design an NVM that can be integrated with CMOS process technology without introducing additional processing steps. These integration efforts have also involved endeavoring to use pFET-based NVM, rather than the more traditional nFETs-based NVM. The reason for this is that pFET-based memory cells exhibit various performance advantages compared to nFET-based memory cells. pFETs have the following advantages over their nFET-based NVM counterparts: 1) increased program/erase cycle endurance (due to reduced oxide wearout); 2) availability in logic CMOS processes (due to reduced memory leakage arising from more favorable oxide physics); 3) ability to easily store analog as well as digital values (due to precise memory writes); and 4) smaller on-chip charge pumps (due to decreased charge-pump current requirements).
Whether using pFETs or nFETs as NVM transistors, the possibility of “overtunneling” in either type of such cells poses a significant problem. The referred to “overtunneling” problem, in pFET-based memory cells for example, manifests itself as follows. Such cells use electron tunneling to raise the floating-gate voltage, and impactionized hot-electron injection (IHEI) to lower the floating-gate voltage. One characteristic of the IHEI programming method is that the MOSFET channel must be conducting current to allow electrons to inject onto the floating gate. If during a prior tunneling cycle the floating-gate voltage was raised so high that the pFET was turned off, there will be no channel current when a write to the cell is attempted. Effectively, by overtunneling the memory cell, the memory cell becomes “stuck” in an off state, and in the absence of channel current no electron injection can be performed during a programming (i.e. injection) cycle to lower the floating-gate voltage.
The overtunneling problem, at least as observed in pFET-based NVM cells, detracts from their use as reliable memory devices. Accordingly, there is a need for methods and apparatuses for preventing overtunneling in NVM cells.